How much does AI voice cloning actually cost?

October 22, 2026 · 6 min read · By Resona Editorial Team · Resona Blog

"Voice AI costs $X" is a meaningless sentence until you know the billing model. A $15/month subscription and a pay-per-byte plan can differ by an order of magnitude for the same project — in either direction. This article breaks down the three models you will encounter, with worked numbers.

The three pricing models

Subscription. You pay a flat monthly fee for a quota of characters or minutes. Simple to budget, painful in practice: you pay for capacity you do not use, quotas reset monthly, and the cheapest tier usually excludes the features you actually want (commercial rights, cloning, higher quality models).

Pay-as-you-go credits. You top up a balance and every action deducts from it. No monthly fee, no quota reset, and — on the better platforms — credits never expire. You pay only for what you generate.

Per-unit metered billing. Metered by character, byte, or second of audio. Technically the most precise, but it is designed for developers integrating speech into a product; the unit prices are meaningless to a creator who just wants a narration.

Worked example: a ten-minute narration

Suppose you need a single ten-minute voiceover — say, 1,500 words of English, roughly 9,000 characters. Here is what each model charges for the same output:

Subscription: one month at $15–$30 covers it easily. Your effective cost for this project: $15–$30, whether you generate ten minutes or ten seconds. Generate nothing next month? Still $15–$30.
Pay-as-you-go (Resona's model): text-to-speech is billed at 4,200 credits per million UTF-8 bytes. 9,000 bytes × (4,200 / 1,000,000) ≈ 37.8 credits. At $1 = 50 credits, that is about $0.76. Plus a one-time voice model: 100 credits ≈ $2. Total for the project: about $2.76, and the voice model stays reusable.

Now the honest counter-example: if you generate twenty hours of audio every month, the subscription wins on pure unit economics. Usage-based pricing is not automatically cheaper — it is automatically fairer, because low usage costs little and heavy usage costs proportionally.

The per-byte subtlety (why Chinese "costs more")

Byte-based billing counts the text you send, not the audio you get. English averages about one byte per character; Chinese, Japanese and Korean characters take about three bytes each in UTF-8. So the same sentence costs roughly three times more in CJK than in English.

The saving grace: a minute of spoken Chinese contains roughly a third as many characters as a minute of English. Per minute of finished audio, CJK and English land in a similar range — the difference only shows up if you compare character counts instead of runtime.

The other cost lines people forget

Cloning fee. Most platforms charge per voice model created. On Resona that is 100 credits (about $2) per clone — once per voice, not per generation.
Transcription. Speech-to-text is usually billed per second of audio. On Resona: 0.2 credits per second, so one hour of audio is 720 credits ≈ $14.40.
Voice design. Generating a voice from a text description is billed per request — 5 credits per successful request on Resona.
Credit expiry. Ask before you buy. Some platforms void unused credits after 12 months; ours never expire. A balance that quietly evaporates is a price increase you did not agree to.
Feature gates. The cheap tier often excludes commercial use or the better models. Check what the tier you are comparing actually includes.

How to estimate your own project

Three numbers answer most questions:

How many minutes of finished audio do you need per month?
How many distinct voices will you clone (one-time cost each)?
Do you need transcription too, and how many hours of audio?

Multiply minutes by ~9,000 bytes per ten minutes of English, apply the per-byte rate, add the clone fees — you have your monthly bill. If it stays under a subscription's price for three consecutive months, usage-based pricing is the right fit; if it consistently exceeds it, a subscription with a quota you actually use is fine too.

Where Resona sits

Full transparency: Resona has no monthly subscription and no starter allowance. You buy credit packs starting at $9 ($1 = 50 credits), credits never expire, and every deduction is itemized in your dashboard. Cloning is 100 credits, TTS is 4,200 credits per million UTF-8 bytes, transcription is 0.2 credits per second. The complete rate card is on the pricing page — no usage fine print.

To see these rates applied to your own script, open the Resona console: the estimator shows the credit cost before you generate. Credit packs start at $9 and credits never expire — full rate card on the pricing page.

New to this topic? Read the full Voice Cloning Guide.